Are you sure?
This action might not be possible to undo. Are you sure you want to continue?
r
X
i
v
:
g
r

q
c
/
0
3
0
1
0
4
2
v
2
2
8
M
a
r
2
0
0
3
Inﬂation without a beginning: a null boundary proposal
Anthony Aguirre
School of Natural Sciences, Institute for Advanced Study Princeton, New Jersey 08540, USA
∗
Steven Gratton
Joseph Henry Laboratories, Princeton University, Princeton, New Jersey 08544, USA
†
(Dated: February 7, 2008)
We develop our recent suggestion that inﬂation may be made past eternal, so that there is no initial
cosmological singularity or “beginning of time”. Inﬂation with multiple vacua generically approaches
a steadystate statistical distribution of regions at these vacua, and our model follows directly from
making this distribution hold at all times. We ﬁnd that this corresponds (at the semiclassical level)
to particularly simple cosmological boundary conditions on an inﬁnite null surface near which the
spacetime looks de Sitter. The model admits an interesting arrow of time that is welldeﬁned and
consistent for all physical observers that can communicate, even while the statistical description
of the entire universe admits a symmetry that includes timereversal. Our model suggests, but
does not require, the identiﬁcation of antipodal points on the manifold. The resulting “elliptic” de
Sitter spacetime has interesting classical and quantum properties. The proposal may be generalized
to other inﬂationary potentials, or to boundary conditions that give semieternal but nonsingular
cosmologies.
PACS numbers: 98.80.Cq, 04.20.Gz, 04.62.+v
I. INTRODUCTION
The ancient philosophical question of whether the uni
verse is ﬁnite or inﬁnite in time, and whether time “had
a beginning”, entered the domain of scientiﬁc study dur
ing the 20th century with the development of generally
relativistic cosmologies in which the universe is homoge
neous and expanding, as observed on the largest accessi
ble scales.
Chief among these cosmologies, and representing oppo
site answers to the question of the universe’s beginning,
were the classical BigBang and SteadyState. As FRW
models, both are based on some form of the “Cosmolog
ical Principle” (CP) that the largescale statistical prop
erties of the universe admit spatial translational and ro
tational symmetries. The models diﬀer greatly, however,
in their time evolution. In the BigBang, the properties
of the universe evolve in a ﬁnite time from a dense, singu
lar initial state. In contrast, the SteadyState universe is
said to obey the “Perfect Cosmological Principle” (PCP)
in that it admits, in additional to spatial translational
and rotational symmetries, a timetranslation symmetry.
Since all times are equivalent, there can be no “beginning
of time”, and the universe is inﬁnite in duration.
Unlike their philosophical predecessors, the BigBang
and SteadyState model were observationally distinguish
able, and astronomical evidence eventually turned nearly
all cosmologists away from the SteadyState. Moreover,
theorems proven within General Relativity showed that
the classical singularity of the BigBang cosmology was
∗
Electronic address: aguirre@ias.edu
†
Electronic address: sgratton@princeton.edu
robust and could not be avoided by relaxing simplify
ing assumptions such as that of homogeneity. Thus the
idea of a temporally ﬁnite universe with a singular initial
epoch came to dominate cosmology.
1
Attention has since
focused on how this primordial singularity (where some
presently unknown theory of quantum gravity presum
ably applies) could give rise to a classical “initial” state
that could evolve into the observed universe.
The required “initial” classical state, however, seemed
rather special: the universe had to have been extremely
ﬂat, and statistically homogeneous (i.e. obey the CP)
on scales larger than the horizon size. The theory of
inﬂation was devised and widely accepted as a solution
to this problem of a special initial state: given inﬂation,
a ﬂat, homogeneous universe (with the necessary Gaus
sian scaleinvariant density ﬂuctuations) is an attractor.
That is, within some inﬂating region of ﬁxed, ﬁnite phys
ical size, the CP holds more and more precisely with
time. What is perhaps more surprising and less widely
appreciated, however, is that in generic inﬂation models
the universe also comes to obey, with evergreater preci
sion, the Perfect Cosmological Principle. This occurs be
cause inﬂation is generically “semieternal”: rather than
ending globally at some time, inﬂation always continues
in some regions, and the universe globally approaches a
quasisteadystate distribution of inﬂating and thermal
ized regions, the statistical description of which becomes
1
There have been a number of proposals for avoiding a beginning
of time, but generally these involve either continuing through
the cosmological singularity by invoking quantum gravitational
eﬀects, or modifying GR at the classical level. Our approach aims
to develop a nonsingular cosmology without appeal to either
possibility.
2
asymptotically independent of time [1].
Since inﬂation generically approaches a steadystate, it
seems physically reasonable to ask whether the universe
can simply be in an inﬂationary steadystate, thus avoid
ing a cosmological singularity or “beginning of time”. In
deed, the possibility of truly eternal inﬂation was raised
soon after inﬂation’s invention, but no satisfactory model
was immediately devised [2], and in subsequent years
several theorems were formulated proving the geodesic
incompleteness of models globally satisfying conditions
seeming necessary for eternal inﬂation [3, 4]. These the
orems suggested that inﬂationary cosmologies necessarily
contain singularities, but the exact nature of the implied
singularities was obscure.
In a recent paper, we constructed a counterexample
to these theorems by providing a model for geodesically
complete truly eternal inﬂation [5]. There, we analyzed
the classical SteadyState model in detail, then extended
our analysis to inﬂation. Here, we develop the model
from a diﬀerent standpoint, focusing on an eternal inﬂa
tion in a doublewell inﬂaton potential, and on the cor
responding cosmological boundary conditions. Section II
motivates and develops our model, and describes its gen
eral features. Various aspects of the model are developed
in subsequent sections: Section III discusses the arrow of
time in our model, and elucidates the failure of the sin
gularity theorems to forbid our construction; Sec. IV dis
cusses the cosmological boundary conditions, which are
speciﬁed on a null surface; Sec. V discusses the relation
of our model to the “antipodally identiﬁed” or “elliptic”
interpretation of de Sitter spacetime that it suggests, and
brieﬂy discusses quantum ﬁeld theory in elliptic de Sit
ter; Sec. VI discusses generalizations and extensions of
our model. We summarize and conclude in Sec. VII.
II. THE PROPOSAL
In this section we develop an eternally inﬂating cos
mology based on a doublewell inﬂaton potential U(φ)
with minima at φ
t
and φ
f
, where U(φ
f
) > U(φ
t
) ≥ 0.
This sort of potential is posited in “old” inﬂation [6] or
“open” inﬂation [7, 8]. We will ﬁrst review semieternal
doublewell inﬂation, then extend this to eternal inﬂa
tion, then analyze and address the geodesic completeness
of the model.
A. Semieternal “doublewell” inﬂation
A semieternally inﬂating cosmology naturally arises
from a generic doublewell potential [9, 10]. Consider
some large comoving region in which, at an initial time
t = t
0
, the energy density is dominated by the inﬂaton
φ, with φ ≃ φ
f
and ∂
µ
φ ≃ 0. Assume the spacetime to
locally resemble de Sitter (“dS”) space, with (for conve
nience) nearlyﬂat spatial sections, i.e. with a metric [11]
approximated by
ds
2
= −dt
2
+e
2Ht
dx
2
+dy
2
+dz
2
. (1)
In this background, bubbles of true vacuumφ
t
nucleate
at a ﬁxed rate λ per unit physical 4volume that depends
upon the potential U(φ)[12]. The interior of each bubble
looks like an open FRW cosmology to observers inside it.
For a suitably designed U(φ) (as in open inﬂation[7, 8]),
there can be a slowroll inﬂation epoch inside the bub
ble so that the FRW regions are nearly ﬂat and homo
geneous, and have scaleinvariant density perturbations.
One such region could therefore in principle represent our
observable cosmological surroundings.
At any time t, we can derive the distribution of bub
bles and inﬂating region within our comoving volume,
with the aim of showing that the distribution approaches
a steadystate. To avoid complications resulting from
bubble collisions and the ambiguities in connecting the
timeslicings within and outside bubbles, we concentrate
on the statistics describing the inﬂating region outside
of the bubbles. This region is necessarily unaﬀected by
the bubbles’ presence because they expand at the speed
of light: both its global and local properties depend only
on its initial state at t
0
. But we may describe three eﬀects
of the bubble encroachment upon it.
First, let us consider the inﬂating region left at time
t by bubbles forming since t
0
. This region must consist
entirely of points each of which does not have a nucle
ation event in its past lightcone (PLC) going back to t
0
.
Denoting the volume of this PLC by Q(t, t
0
), it can then
be shown that such points comprise a volume fraction
f
inf
= exp[−λQ] ≃ exp
¸
−4πλ(t −t
0
)
3H
3
(2)
for (t − t
0
) ≫ H
−1
[9, 10]. Although f
inf
→ 0 for large
t − t
0
, the spacetime is said to be eternally inﬂating be
cause for small λ the physical inﬂating volume within our
comoving region nonetheless increases exponentially with
time:
V
inf
∝ f
inf
exp(3Ht) ∼ exp(DHt) (3)
for any ﬁxed t
0
, with D ≡ 3 −4πλ/3H
4
.
Second, one can show [9] that at ﬁxed t the distribution
of inﬂating regions about any inﬂating point is described
by a fractal of dimension D (that is, the inﬂating volume
V
inf
∝ r
3
f
inf
(r) ∝ r
D
) up to a scale of order r
B
(t, t
0
),
where
r
B
(t, t
0
) = H
−1
e
H(t−t0)
−1
(4)
is the physical radius at t of a bubble nucleated at t
0
.
Third, we may calculate, for a given point in the in
ﬂating region at time, the number per unit time N(r, t)
of incoming bubbles of physical radius r. This is
N(r, t) =
4πλr
2
(1 +Hr)
4
(5)
3
II
I
i
0
J

J

J
+
i
0
Q
P
X
Y
x
i
FIG. 1: Conformal diagram for de Sitter (dS) spacetime.
Each point represents one point in 1+1 D dS, or half of a
2sphere in 3+1 D dS. The left and right (dotted) edges are
identiﬁed. The shaded region (region I) is covered by coor
dinates with ﬂat spatial sections (spacelike lines) with space
like inﬁnity at i0; the straight, timelike lines represent co
moving geodesics. The null surface J
−
represents t → −∞.
Truevacuum “test bubbles” (not disturbing the background
spacetime) are darkly shaded and open toward future time
like inﬁnity J
+
. Also shown are are the lightcones of points
P and Q in regions I and II that open toward J
−
, and null
(“Y”) and timelike (“X”) geodesic segments with an endpoint
at xi and crossing J
−
.
for r < r
B
(t, t
0
) and zero for r > r
B
.
An observer within a bubble can never leave, but will
eventually be encountered by an encroaching bubble wall
after a typical time τ
coll
, where τ
−1
coll
is related to the r
integral of Eq. (5) by some transformation between the
bubble observer’s proper time τ and cosmic time t. Since
this rate depends on t − t
0
, a patient and very sturdy
observer could in principle discover the global time at
which it formed by counting the frequency of incoming
bubbles.
Now, as t → ∞, four things occur. First, the nearly
ﬂat spatial sections approach perfect ﬂatness. Second,
the incoming bubble rate N(r, t) becomes homogeneous
and independent of time on arbitrarily large physical
scales. To see this, imagine that the rate is inhomo
geneous at early times because bubble nucleation starts
at diﬀerent times in diﬀerent regions. But since the im
pact rate depends on the initial time only for bubbles
of radius greater than r
B
(t, t
0
) ∼ exp[H(t − t
0
)], it is
then homogeneous and independent of time on arbitrar
ily large scales as t → ∞. Third, and for essentially the
same reason, the distribution of inﬂating region around
any given inﬂating region also becomes homogeneous and
independent of time on arbitrarily large scales. Fourth,
observers within bubbles lose the information about the
“global time” at which they exist (see Eq. (5)), and all
bubbles become equivalent. Thus the physical descrip
tion of the universe, relative to any ﬁxed length scale
such as H
−1
, satisﬁes the Perfect Cosmological Principle
arbitrarily well as t →∞.
B. Eliminating the beginning
The above semieternally inﬂating model can be made
eternal by setting the “state” of the universe to be exactly
that state approached by semieternal inﬂation: because
the statistical properties depend only upon t − t
0
, for
speciﬁed conditions at t
0
the state at ﬁxed t with t
0
→
−∞ is the same as that for t →∞ with ﬁxed t
0
.
The state so obtained has the four basic character
istics listed above: The spatial sections are exactly ﬂat
(outside of the bubbles), the bubble distribution (as char
acterized by the incoming bubble rate) is homogeneous
and independent of time, as is the distribution of inﬂat
ing regions about any inﬂating region, and the bubbles
are all statistically identical. The inﬂating region is a
fractal of dimension D < 3 on all scales. This means
that, although inﬂating regions exist, the global inﬂating
fraction f
inf
is zero, just as the fraction of 3D Minkowski
spacetime ﬁlled by an inﬁnite 2plane of ﬁnite thickness—
an object of fractal dimension two—would vanish. (The
zero probability that a randomly chosen point is in an
inﬂating region accords with the fact [5] that within the
PLC of each point there is an inﬁnite 4volume in which
bubbles can nucleate toward that point.) Unlike the re
gion ﬁlled by the plane, however, the inﬂating region is
statistically homogeneous and isotropic, in that it ex
actly satisﬁes the “Conditional Cosmographic Principle”
of Mandelbrot that the statistical description of the in
ﬂating region about any given inﬂating region is inde
pendent of the inﬂating region chosen (see Ref. [13] for
a discussion of this and other aspects of “cosmological”
fractals).
This model (essentially derived by Vilenkin [9]) would
seem to have exactly the properties expected of an eter
nally inﬂating spacetime, has been straightforwardly con
structed using the steadystate generated by a semi
eternally inﬂating model, and extends to inﬁnite nega
tive cosmic time. Yet the arguments of [3, 4] (discussed
in more detail in Sec. III) imply that it should be geodesi
cally incomplete. This issue can be addressed with ref
erence to the conformal diagram of the model, shown
in Fig. 1. The background inﬂating spacetime is rep
resented by the lightly shaded region. Each equaltime
surface is intersected by an inﬁnite number of bubbles
(indicated by lightcones opening toward J
+
), which are
concentrated along J
+
and near i
0
.
The model thus far constructed (deﬁned in the shaded
region henceforth called “region I”) is geodesically in
complete: all null geodesics (such as “Y” in Fig. 1), and
all timelike geodesics (such as “X”) other than the co
moving ones, have only a ﬁnite proper time (or aﬃne
parameter) between a point in region I and coordinate
time t →−∞ (see, e.g., [5]). Most geodesics thus “leave
the spacetime” to their past, encountering J
−
, the limit
surface of the ﬂat equaltime surfaces as t → −∞. (On
the conformal diagram this surface looks like a null cone
emanating from a point at the bottom edge.)
Although the spacetime is geodesically incomplete
4
there is no curvature blowup or other obvious pathol
ogy at J
−
, so the spacetime is extendible rather than
singular. One may take the position that this sort of in
completeness is allowed, since the edge is outside of the
future of any point in the region, and any given thing in
the spacetime was made at some particular coordinate
time t > −∞
2
. From this point of view, there is no clear
reason to reject the model as deﬁned in region I.
It seems quite reasonable, however, to ask instead how
the manifold could be extended, and what could be in the
extension. We start by extending the manifold to include
J
−
, which is the boundary of the open set comprising
region I. We shall see, as follows, that on J
−
the ﬁeld
must everywhere be in the false vacuum. Deﬁne φ(λ) as
the ﬁeld value at aﬃne parameter λ of a noncomoving
geodesic starting at some arbitrary point x
i
in region I,
where λ increases away from J
−
. We know that for
some aﬃne parameter λ
J
, the geodesic encounters J
−
,
and also that if our point is within a bubble, there is also
a ﬁnite value λ
f
> λ
J
at which the geodesic leaves the
bubble and enters the false vacuum φ
f
. Then for λ < λ
f
,
we have
lim
λ→λJ
φ(λ) = φ
f
. (6)
If we then require that the ﬁeld be continuous along any
geodesic, we then ﬁnd that the ﬁeld must be in the false
vacuum φ
f
everywhere on the surface J
−
. That is, at
the semiclassical level of description, J
−
must be an in
ﬁnite null surface of pure false vacuum, through which
no bubbles pass.
Let us now examine the global classical structure of the
background spacetime by momentarily neglecting semi
classical processes such as bubble nucleations. Then the
manifold comprised of region I and J
−
is locally dS (con
stant Ricci scalar R and vanishing Weyl curvature tensor)
everywhere, with φ in the false vacuum (φ = φ
f
). This
manifold can still be extended past J
−
, and the obvious
extension is to complete dS spacetime. This is certainly
a solution compatible with our state at J
−
, and (as we
will argue in Sec. IV and the Appendix) it seems likely to
be unique. Thus we will take the maximal extension of
the background spacetime to be full dS spacetime. That
is, the nonshaded region of Fig. 1, henceforth called “re
gion II” must simply be the rest of dS spacetime. Con
sider now a classical ﬁeld in the background spacetime
obeying a homogeneous hyperbolic equation. Given any
point P in region I, almost all inextendible nonspacelike
curves through P intersect J
−
. Therefore specifying the
ﬁeld values on J
−
eﬀectively poses a “characteristic ini
tial value problem” [15, 16] with a unique solution ev
erywhere in region I (this is the analog of the Cauchy
2
This was the view taken by investigators of the “cyclic
model” [14] which is geodesically incomplete in a very similar
way, and may be the view taken by the adherents of the classic
steadystate model.
II
I
J

J

P
P
F
P
P
F
X
?
?
t=const.
V

V
+
FIG. 2: Conformal diagram for eternal doublewell inﬂation.
Bubbles are open FRW regions; equal time slices shown as
curved horizontal lines. For clarity we have not included bub
ble intersections. Also shown are past lightcones, cut oﬀ at
J
−
, of both a point P and its antipode −P (note that P
and −P are also reﬂected across the suppressed twospheres
in the 4D case).
problem, but with boundary conditions on a null surface;
see Sec. IV and the Appendix for more details). Exactly
the same argument can be made, however, for any point
in region II. Thus specifying classical ﬁelds everywhere
on J
−
determines their values everywhere in dS space
time. This means that the conditions found to obtain
on J
−
(by specifying the state in region I and requiring
ﬁelds to be continuous) also determine the state in re
gion II and we can extend our model to region II in an
essentially unique way.
We may now examine the extension of the model to
region II at the semiclassical level by including the bub
ble nucleations. The form of U(φ) indicates that bubbles
must nucleate
3
at a ﬁxed rate per unit physical 4volume.
In region I, this led to an asymptotically steadystate
bubble distribution which, when made exact, implied
that there are no bubbles passing through J
−
. Thus in
region II, though bubbles must nucleate at the required
rate, none must pass through J
−
.
The only way this may occur is, in exact symmetry
with region I, to have a steadystate bubble distribution
on the ﬂat slices of region II, with the bubbles opening
away from J
−
. This is illustrated in Fig. 2. Region II is,
3
Bubble nucleation has been perhaps most rigorously analyzed in
Ref. [17], and the boundary conditions for bubble nucleation in
our model correspond exactly to those analyzed in their study.
5
then, a sort of mirror reﬂection of region I through J
−
.
Thus the answer to the question of how the model can be
extended, and what lies beyond J
−
, is that an essentially
identical copy of region I lies in region II, connected by
the inﬁnite null surface J
−
.
This completes the basic speciﬁcation of our model.
The cosmology obeys the CP and PCP in that it ad
mits a coordinatization such that all spatial slices are
statistically homogeneous, and statistically the same as
all others. There is no preferred time in this slicing, nor
is there a cosmological singularity: the model is geodesi
cally complete. Given the appropriate inﬂaton potential,
any one of the bubbles could describe our observable sur
roundings.
III. THE ARROW OF TIME
If we consider all bubbles to expand with time, then
Fig. 2 suggests that while in region I the future is to
ward the top of the diagram (“up”), in region II future
lies toward the bottom (“down”). This leads us to the
issue of the cosmological arrow of time (AOT): why does
the timeasymmetric 2nd law of thermodynamics hold
universally, given that fundamental physics is thought to
admit a symmetry (CPT) that includes timereversal?
There is some consensus that if this question has an an
swer, it must ultimately be cosmological, with the time
asymmetry resulting from some qualitative diﬀerence be
tween cosmological “initial” and “ﬁnal” conditions [18]
that precludes a timereversal (T) symmetry of the phys
ical state in any subregion of the universe, and hence
induces an AOT.
Although our our model has no “initial” conditions, it
does have boundary conditions on J
−
(discussed in de
tail in Sec IV below) and we can discuss the AOT in light
of them. To do so we must divide the universe into two
types of subregions: those entirely outside of bubbles,
and those partially or wholely within them. Outside of
the bubbles (or alternatively near enough to J
−
) there
is no local AOT: the description of such a region admits a
timereversal symmetry. Were we to hypothesize, for ex
ample, an imaginary observer outside of a bubble with its
own AOT (pointing away from the time of its creation),
that observer would see only Tinvariant dS spacetime.
The observer could not know whether it moved “up” or
“down” on the conformal diagram, nor if it was in region
I or II, nor if it crossed J
−
. What the observer is guar
anteed, however, is that it will eventually be encountered
by, and ﬁnd itself within, a bubble.
The bubble interiors are not timesymmetric: within
a bubble, there is a unique time direction in which the
mean energy density decreases. This direction is away
from the bubble wall/slowroll inﬂation epoch, at which
the FRWregion is known to be nearly homogeneous. If
one bubble is to represent our observable surroundings,
this direction must correspond to the time direction in
which the entropy of an isolated system increases. It
}
∆=const.
O
t=const.
φ=const.
t
x
V(x)
FIG. 3: The decay of: (left) an unstable particle, (right) an
unstable vacuum state. In the particle case, the trajectory
is classically describable to good precision at early and late
times, but not near the decay (shaded region). Likewise, the
inﬂaton φ is classically describable at large invariant distance
∆ from the nucleation event at O, but not near it (shaded re
gion). This quantum region connects the classically describ
able ﬁeld conﬁguration of the bubble interior to that of locally
dS spacetime.
has been often argued, particularly by Penrose [19], that
this connection arises because when gravity is included
homogeneity corresponds to an extremely lowentropy
state. We shall assume this correspondence here (and
that the bubble does not begin in some very special state
for which the density ﬂuctuations decrease). Under this
assumption the physical AOT within any bubble must
point away from the bubble walls; globally this means
that the AOT (where deﬁned) points away from J
−
.
As illustrated in Fig. 2, one can therefore indeed draw
timelike geodesics (such as “X”) along which the physical
AOT reverses, but the reversal always occurs to the past
of any physical observer (all of which are within bubbles),
and within a region (the locally dS spacetime) in which
there is no welldeﬁned physical AOT.
We have argued that within bubbles the physical laws,
but not the physical state, admit a symmetry (CPT) in
cluding timereversal, while outside of bubbles the laws
and the state admit time symmetry; but what about the
bubble nucleations themselves? Is there not some AOT
telling them “which way” to nucleate, depending upon
which side of J
−
they are on?
To clarify this point, consider the process of semi
classical bubble nucleation by analogy to the decay of
an unstable particle, described by a potential V (x) as in
Fig. 3. Classically, one allowed trajectory is for the parti
cle to sit at constant x in the false minimum x
f
(vertical
line on diagram); a second, equalenergy trajectory is
for the particle to “bounce” oﬀ of the potential (curved
line in diagram). A purely quantum description of the
system (in the Schr¨odinger representation) would start,
for example, with a Gaussian wave function centered at
x
f
at some time t = 0, and hence with the position ex
pectation value ˆ x ≃ x
f
. With time, the wave function
6
spreads, and ˆ x increases, eventually approaching the
classical trajectory for ˆ x ≫ x
f
(we could say roughly
that the particle has “decayed” when ˆ x changes sig
niﬁcantly from x
f
.) Note also that the same spreading
would occur at t < 0 so that this purely quantum de
scription would be Tsymmetric. Now, a semiclassical
description of the system would describe the system clas
sically at both early and late times, but with a quantum
mechanical transition connecting the classical trajecto
ries at some given time. This transition time is random,
and in an ensemble of such systems (as is required for
a correct probabilistic description) would follow a prob
ability distribution given by a WKBtype calculation of
the decay rate. Near the transition time the system can
not be described classically; we must “shade out” the
region where only a quantum description is accurate, as
in Fig. 3.
Bubble nucleation can be described semiclassically in
a similar way. Here, we must attach allowed classical so
lutions of the ﬁeld equations along some boundary that
represents the nucleation event. To do this in a covari
ant way, this boundary must be a surface of zero proper
distance, i.e. a null cone, as shown in Fig 3. A bubble nu
cleation “event” is thus comprised of a region (shown as
the shaded upper quadrant) where a classical bubble in
terior solution applies, attached to locally dS regions by
a “shaded out” region (within some proper separation
squared of the nucleation point) where only a quantum
description is valid. To produce a semiclassical descrip
tion of a spacetime in which nucleation events occur, one
must then populate it with these conﬁgurations in such a
way that the nucleation sites are randomly situated and
occur at the correct rate per unit fourvolume, while the
classical description, which applies far from the nucle
ation site, is in accord with the (classical) boundary con
ditions. When the classical boundary conditions are ours,
given on J
−
, this yields the bubble distribution indi
cated in Fig. 2. The (semiclassical) boundary conditions
do, then, control the time direction of bubble nucleation,
not by introducing some locallydetectable AOT, but by
controlling the allowed global conﬁguration of bubble nu
cleation events.
There is one ﬁnal “region” in which we can check
the AOT: the entire universe. Interestingly, we here
ﬁnd that while each bubble nucleation event is nontime
symmetric, by virtue of the symmetry of the cosmologi
cal boundary conditions, the statistical description of the
universe does admit a sort of Tsymmetry. In Sec. V we
will discuss the possibility of making this symmetry ex
act via an identiﬁcation on the manifold. This raises the
intriguing possibility of having a welldeﬁned (and con
sistent among communicating observers) AOT for all ob
servers even while the physical laws and the global physi
cal description of the universe both admit a timereversal
symmetry.
A. The singularity theorems
Having examined the AOT, we may turn to the sin
gularity theorems, for the AOT proves crucial in their
analysis. Both the older theorems [3] and the newer theo
rem [4] assert that if certain conditions are satisﬁed every
where within a spacetime, then not all past nonspacelike
geodesics have inﬁnite proper time or aﬃne parameter.
This indicates that the spacetime is either extendible or
contains singularities.
The older theorem poses four such conditions, of which
our model satisﬁes three, as does region I or II by itself.
The fourth condition is that there exists a point P such
that there is a ﬁnite diﬀerence in volume between the
interior of the PLC of P and that of any point P
′
in
the past of P. This is motivated by doublewell inﬂa
tion, and claimed as necessary for semieternal inﬂation
(the argument being that if this condition does not hold,
then any inﬂating point will ﬁnd itself in a bubble at
the next instant, with probability one). We argue that
this condition is not quite necessary: what is required is
that there be a ﬁnite volume in the region between the
PLCs of P and P
′
in which bubbles can nucleate toward
P. Thus if, as in the proof of the theorem, the“past of P”
is taken to be the full volume interior to the lightcone
pointing in the time direction away from which a bubble
nucleated at P would expand, then the fourth condition
applies to region I alone (and correctly implies that it is
extendible). But it would not apply to the full spacetime
(which is neither extendible nor singular), because the
relevant part of the lightcone extends only to J
−
.
The argument of the newer singularity theorem [4] con
sists of the deﬁnition of a local “Hubble parameter” H
meant to represent the rate of divergence of neighbor
ing comoving test particles, along with an argument that
any region in which a suitable average H
av
of H is greater
than zero along all geodesics must be geodesically incom
plete.
We understand Borde at al.’s argument as follows. One
imagines some timelike or null test geodesic with aﬃne
parameter λ in the spacetime in question, then attempts
to construct a timelike vector ﬁeld u
µ
(λ) along the test
geodesic into its past such that H
av
(deﬁned via u
µ
) ex
ceeds zero. It is shown that this can only be achieved
along some ﬁnite aﬃne length of the test geodesic, since
the imposed condition rapidly forces u
µ
towards nullness.
Borde et al. [4] then take their result to mean that an
eternally inﬂating spacetime is pastgeodesically incom
plete.
We take the hypothetical satisfaction of their averaged
Hubble parameter condition for all test geodesics as the
implicit deﬁnition of what Borde et al. mean by an eter
nally inﬂating spacetime. The logic is that u
µ
might be
independent of test geodesic and simply be the velocity
ﬁeld of some set of comoving worldlines in the inﬂating
spacetime. So what the theorem actually implies is that
is it impossible to entirely cover a spacetime with such a
set of worldlines in a way that allows all test geodesics
7
cutting these worldlines to obey the Hubble parameter
condition.
For illustration, let us consider dS space. We note that
as pure dS is a maximally symmetric spacetime, it does
not make sense to regard some parts as inﬂating and oth
ers not—one must break the symmetry by adding some
other ingredients, and then frame a discussion in terms
of them. Nevertheless, let us investigate if the geodesics
deﬁned by having ﬁxed comoving coordinates themselves
constitute a u
µ
ﬁeld satisfying H
av
> 0. It turns out
that H reduces to the usual ˙ a/a in this situation. In the
closed slicing with metric
ds
2
= −d
˜
t
2
+H
−2
cosh
2
H
˜
t
dχ
2
+ sin
2
χdΩ
2
2
, (7)
the coordinates cover all of dS, so u
µ
is globally deﬁned.
However, H < 0 for
˜
t ≤ 0 so H
av
goes negative there.
Now consider using a single ﬂat or open coordinate patch
to deﬁne u
µ
. H
av
> 0 here (at least for the appropri
ate choice of time orientation). However, because nei
ther of these coordinate patches covers the spacetime,
neither does u
µ
. Furthermore, one may choose an inﬁ
nite number of ﬂat coordinatizations of dS, each with a
diﬀerent null boundary where the u
µ
construction fails.
This makes it clear that the boundary of a given u
µ
ﬁeld
cannot be unambiguously used to deﬁne an edge to an
inﬂating region.
We do not believe that the global existence of a suitable
u
µ
is necessarily the best deﬁnition of what is meant by
an eternally inﬂating spacetime.
4
In particular, equat
ing H with physical expansion entails a tacit assumption
that the physical AOT is everywhere in accord with that
deﬁned by u
µ
. The model we have proposed could be
covered by a congruence of geodesics (those comoving in
two ﬂat coordinate patches covering dS) that would yield
H < 0 in some regions. However, we have argued that
these regions may still be regarded as expanding with re
spect to the physical AOT deﬁned by the cosmological
boundary conditions.
In summary, both singularity theorems postulate con
ditions for a region to be “inﬂating”, and ﬁnd that such
a region cannot be geodesically complete. However, in
terpreting these theorems as forbidding eternal inﬂation
seem to us to require an unwarranted assumption about
the global AOT independent of the cosmological bound
ary conditions.
4
Indeed the term “eternal inﬂation” has been used with a vari
ety of meanings. For example, the recent paper [20] used it to
describe models that are eternally inﬂating to the future, but
simply geodesically complete to the past ( ˙ a and/or ¨ a may go
negative there). Such histories, with a globallydeﬁned arrow of
time, seem physically unrealistic with a typical mechanism for
exiting inﬂation, since this would render them unstable to the
formation of thermalized regions in the putatively eternal early
phase[2].
IV. BOUNDARY CONDITIONS AND THE
NULL BOUNDARY PROPOSAL
We have seen how extending semieternal inﬂation to
eternal inﬂation implies particular behavior on the inﬁ
nite null surface J
−
. Here we discuss the converse, de
scribing how the eternal doublewell inﬂation model we
have described can be speciﬁed by a particularly simple
set of cosmological boundary conditions that are imposed
on J
−
.
A. On cosmological “initial” conditions
The correct speciﬁcation even of a complete set of
physical laws does not by itself allow prediction of any
physical system’s behavior; these laws must be sup
plemented by boundary conditions that suﬃce to fully
characterize the system being modeled. The BigBang
(“BB”) model essentially consists of a set of such bound
ary conditions for our observable cosmic surroundings:
at some early time, our region was a hot, dense, nearly
homogeneous and isotropic mixture of particles and ﬁelds
in thermal and chemical equilibrium, in a nearlyﬂat ex
panding background geometry with scaleinvariant grav
itational potential perturbations of amplitude 10
−5
on
the scale of the cosmological horizon.
While these initial conditions yield predictions in ex
cellent accord with astronomical observations, they are
deemed by many to be too special: they seem to com
prise an extremely small portion of some “ensemble of
all possible initial conditions”, using some (generally un
speciﬁed) measure (see e.g. [21]). The theory of inﬂation
is widely accepted as a way to broaden the range of al
low initial conditions by funneling a relatively wide class
of physical conditions into satisfactory BB initial condi
tions [22].
This is perhaps reasonable as a stopgap measure as
long as the true (preinﬂation) initial conditions are un
known. But it does not solve the initial condition prob
lem, for not all initial conditions will give rise to inﬂa
tion [23], nor is it clear that all of those that do give
inﬂation will yield a viable bigbang model [21]. Thus
one must still assume that suitable conditions emerge
from the initial singularity; whether this assumption cor
responds to a “special” or “generic” condition seems ill
deﬁned without a description of the singularity, which
would presumably require quantum gravity. But unless
quantum gravity completely alters the way physical the
ories are applied, it is unlikely to yield a unique initial
condition; it will still be necessary to supply a quantum
state (or something similar). Even if it is possible to de
ﬁne the ensemble of all possible states, the question of
how (and why) a state was “chosen” for our universe can
only be a metaphysical one, and it seems that, just as in
choosing physical laws, we can at best posit a state on
some (possibly symmetrical or aesthetic or philosophical)
grounds, derive its consequences, and compare them to
8
the observed universe.
The same holds for a nonsingular cosmological model
such as that proposed in this paper, the basic structure
of which is welldescribed by semiclassical physics, and
where there can be no hiding of the cosmological bound
ary conditions in an initial singularity. We instead spec
ify a particular set, and analyze them in terms of their
simplicity, symmetry, and ability to correctly generate
a bigbang like region that can describe our observable
surroundings.
B. (Semi)classical boundary conditions for eternal
inﬂation
It is conventional to pose boundary conditions for a
set of classical ﬁelds by specifying the ﬁelds (and gener
ally their time derivatives) on a spacelike surface. This
accords with the intuitive idea of specifying the state at
an initial time. An alternative procedure, more relativis
tic in spirit because it does not assume a particular time
coordinate, is to specify boundary conditions on a null
surface such as the lightcone of a point. (See the Ap
pendix and Refs. [15, 16, 24] for treatments of the null
initial value problem.)
Our boundary surface J
−
is a null surface that, when
drawn on the conformal diagram, looks like the lightcone
of a point V
−
at the bottom (i.e. “at” the conventional
past timelike inﬁnity) and the lightcone of a point V
+
at
the top; see Fig.2). As such, one might expect that one
may determine ﬁelds (including the spacetime metric)
throughout the space by specifying them on J
−
. The
conformal mapping, however, hides the fact that J
−
is
not a fully closed lightcone: there is always a nonzero
physical volume (of order H
−3
or greater) on any space
like surface bounded by J
−
. We must therefore ask if
specifying ﬁelds on J
−
suﬃces to ﬁx them everywhere, or
if there is information that may come “though the hole”
at V
−
and V
+
. In the Appendix, we argue (using the
Green function for a massive scalar ﬁeld in dS) that the
ﬁelds at V
±
are irrelevant (as long as they are reasonably
wellbehaved), as their eﬀect is inﬁnitely diluted.
We can thus pose boundary conditions for our cosmol
ogy at the classical level as:
1. There exists an inﬁnite connected null hypersurface
J
−
of topology R×S
2
, on which the 4dimensional
Weyl tensor vanishes and the 4D curvature scalar
is constant.
2. The inﬂaton ﬁeld, with a “double well” potential,
is everywhere in the false vacuum on this surface.
3. On this surface, all classical ﬁelds are zero or are
in minima of their potentials. This precludes any
radiation propagating through J
−
.
This cosmology is classically very dull, as it is just
de Sitter space everywhere with no dynamics. However,
semiclassical bubble nucleations can, without aﬀecting
the ﬁelds on J
−
, create interesting dynamics by forming
bubbles that open everywhere away from J
−
, and give
rise to eternal inﬂation as described in Sec. II and shown
in Fig. 2.
C. Quantum mechanical boundary conditions
Although a fully quantum description of semiclassical
phenomena—such as the nucleation of a bubble—is gen
erally prohibitively complex, we may hope that, be
cause our set of boundary conditions is so simple at a
semiclassical level, it might be amenable to a simple
quantum formulation, and it would be pleasing to spec
ify our boundary conditions in an explicitly quantum
mechanical way. There are two ways one might think of
doing this.
First, one might consider doing usual quantum ﬁeld
theory on de Sitter space, but with a particular choice
of vacuum corresponding to our “false vacuum” classical
conditions on J
−
. This might be done by putting the
ﬁeld in a Gaussian wavepacket around the false vacuum
at (ﬂat slicing) time t, as in [17], then taking a t →−∞
limit. One might then assume that this is suﬃcient to
deﬁne the quantum state over all of de Sitter space. Al
ternatively, one could set such an initial state at the
˜
t = 0
slice in global (closed) coordinates (7), i.e. on the slice
through the “throat” of the dS hyperboloid. This surface
could then be boosted (inﬁnitely) to become J
−
. In ei
ther case, the procedure would be analogous to putting
an unstable particle into a Gaussian in the false vacuum
at t = 0; away from t = 0 in both time directions, the so
lution of the timedependent Schr¨odinger equation would
evolve away from the unstable “initial” state.
With our initial data surface being null, one might at
tempt to deﬁne the quantum state directly on it. This
would be conceptually more pleasing, not requiring the
limit through a sequence of spacelike surfaces as above.
Further, a null surface formulation of initial data is rather
elegant [16, 24]. However, unlike for a spacelike initial
value surface, some points on J
−
are in causal contact.
It is thus less easy to see how to move from a spacelike
to a null boundary value surface in QFT, because ﬁelds
do not commute at lightlike separated points. A sim
ilar eﬀect occurs in socalled “lightcone quantization”
approaches to QFT in Minkowski space [25], where one
uses inﬁnite null sheets (or lightfronts) instead of equal
time slices, and ﬁelds again do not in general commute.
The vacuum state for interacting ﬁelds is rather easier
to deﬁne in the lightfront approach to QFT than in the
usual spacelike approach to QFT, and one might expect
this to hold in a proper null cone approach to QFT. Un
fortunately, the authors know of no such formulation of
QFT even in Minkowski space.
Thus while we suspect that our cosmological bound
ary conditions may correspond to a the speciﬁcation
of a rather simple quantum state, we leave this diﬃ
cult problem for future work and here concentrate on
9
a semiclassical description (though we return to QFT in
Sec. VA).
D. Discussion of the null boundary proposal
The boundary conditions we have proposed are ex
tremely simple, in line with the view espoused above that
one cannot avoid making a speciﬁc choice for cosmolog
ical boundary conditions, and that it is then reasonable
to make a choice that is as simple and as highly symmet
ric as possible, rather than hope to choose a “generic”
boundary condition.
1. Relation to other proposals
Our proposal is novel in that it requires no treatment
of an initial singularity, and in that boundary conditions
are placed on a null boundary that does not correspond
to any particular cosmic “time”. It does, however, have
features in common with other proposals for specifying
the state of the universe.
Penrose’s Weyl Curvature Hypothesis (WCH) requires
that initial singularities are constrained to have vanishing
Weyl tensor, corresponding to a lowentropy state. This,
he argues, gives rise to an arrow of time ﬂowing away
from the initial state. Our proposal is very similar, except
of course that we impose the vanishing of the Weyl tensor
on a lightlike slice running across the universe rather than
at an initial singularity.
The HartleHawking No Boundary Proposal (NBP) is
a prescription for the wavefunction of the universe. It
provides Ψ(h
ij
, φ), the amplitude for a given threemetric
and scalar ﬁeld conﬁguration on a spatial 3D surface.
The construction is designed to suppress irregular con
ﬁgurations relative to more regular ones, and to thus fa
vor simple and symmetric states. The notion of temporal
evolution does not appear explicitly. However, one can
associate histories with saddlepoint approximations to
the wavefunction, and it turns out that such universes
are often smooth when they are young and small. Thus
the NBP may be taken to imply that, at the semiclassical
level, there should exist a spatial surface in a cosmological
spacetime on which the boundary conditions are partic
ularly simple.
5
. Our proposal rather uses a null surface
(but see Sec. VI B below). It would be very interesting to
develop a quantum prescription in a similar vein to the
NBP which naturally leads to simplicity on certain null
5
In the context of models with openinﬂationary potentials, the
NBP seems to favor histories in which the scalar ﬁeld is every
where in its true vacuum [26]. The NBP may still be relevant
for inﬂation with the use of an anthropic constraint, or in a
“topdown” approach to calculating quantum probabilities [27].
Starting oﬀ in the true vacuum is however not a problem for
recycling models of inﬂation, as discussed below.
surfaces. For further details on both the WCH and NBP
see [28].
The tunneling approach to quantum cosmology [29] ar
gues that the semiclassical universe emerges via a quan
tum tunneling event. In the context of models with
openinﬂationary potentials the proposal suggests that
the universe, when ﬁrst semiclassically describable, is
most likely to be small and regular, with the ﬁeld away
from the true vacuum; futureeternal inﬂation can then
ensue. The tunneling is supposed to have occurred out
of a quantum gravitational chaos so severe as to pre
clude any spacetime description. While our proposal
also leads to inﬂation from semiclassical boundary con
ditions, it explicitly avoids any such extreme quantum
gravitational regime.
Over the years Sakharov has discussed various cosmo
logical models involving timereversal (and CPT) invari
ance [30, 31, 32, 33]. In these models, the universe is
generally symmetric across a singular FRW bounce at
t = 0, at which universe assumes an especially simple
state; away from this bounce entropy increases in both
directions of time [31]. In [30] he hypothesizes that phe
nomena at t < 0 are the CPT reﬂections of the phenom
ena at t > 0. In [31, 32] he considers the possibility of
an inﬁnite chain of further bounces or cycles away from
t = 0 in both time directions, and also the possibility
that the minimumentropy surface might be one of max
imum expansion rather than a singular one [31, 32]. In a
paper primarily about signature change [33] he alludes to
possibility of nonsingular timereversal around a surface
of minimal radius in a false vacuum state. Our proposal
clearly has parallels with these ideas. We, however, con
centrate on an inﬁnite nonsingular null surface, rather
than a singular spacelike one. Moreover, with the con
cepts of semiclassical bubble nucleation and open inﬂa
tion, we are able to provide a relatively complete physical
picture.
2. The horizon problem
Inﬂation was originally conceived as a remedy for
troubling issues concerning cosmological initial condi
tions [34], thus it is useful to compare how such shortcom
ings of the HBB model [22] are dealt with in our model.
We focus on what is (aside from the singularity problem)
perhaps the most vexing of these HBB diﬃculties: the
horizon problem.
The horizon problem is generally framed as fol
lows: choose two spatially antipodal points on the last
scattering surface. They are similar in temperature, yet
their PLCs never intersect in a HBB cosmology, so there
can be no causal connection between them. Inﬂation is
generally thought to solve this problem, because with suf
ﬁcient inﬂation, the PLCs will intersect. But this does
not suﬃce for the points to have the same temperature,
because the temperature at each point depends on data
across its full PLC, and there is a portion of each PLC
10
that does not intersect the other. Taking this into ac
count, for the two points to have similar properties it
must be assumed that there is suﬃcient inﬂation, and
additionally that at inﬂation’s beginning at time t
inf
, the
region to the past of the two points is homogeneous on
length scales of order H
−1
inf
. But suppose there is a earlier
epoch between some time t and t
inf
. Then at t the patch
must be homogeneous over a physical length scale of at
least
r(t) ∼
a(t)H(t)
a(t
inf
)H(t
inf
)
H
−1
(t), (8)
where H(t) ≡ ˙ a/a. If a ∝ t
α
(α = 1/2 for radiation
dominated expansion) prior to t
inf
, then r/(H
−1
) ≃
[a/a(t
inf
)]
(α−1)
. Thus it is necessary to postulate that
our region was, at some initial time t
0
, homogeneous over
[a(t
inf
)/a(t
0
)]
(3−3α)
Hubble volumes. The horizon prob
lem therefore persists if t
inf
> t
0
and α < 1. There are
two escapes available within inﬂation. The ﬁrst is to set
t
inf
= t
0
= t
pl
, so that the expansion is inﬂationary all
the way back to the Planck time, before which one can
not speak of the expansion at all. The horizon problem
is then greatly ameliorated, as it must only be assumed
that some regions “emerging” from the Planck epoch are
homogeneous over a relatively small (but > 1; see [23])
number of Hubble volumes when they are ﬁrst classically
describable; but whether this constitutes a true solution
will be seen only if and when the Planck epoch itself is
understood. A second potential escape is to set α ≥ 1, or
(more generally) for ˙ a to be nondecreasing, i.e. to have
pasteternal inﬂation.
How, then, does the horizon problem look in the con
text of our eternal model with no initial time? Let us pose
the problem in a slightly more general manner: when
specifying the cosmological boundary conditions, must
one do so over a region that is very large compared to
some relevant physical volume such as H
−3
? Posed this
way, it might seem that because J
−
is an inﬁnite sur
face, a strong horizon problem exists.
6
This, however, is
not necessarily the case. Imagine J
−
as the limit of a se
quence of spacelike slices obtained by boosting the space
like surface given by
˜
t = 0 in global coordinates. Because
each such surface has volume 2π
2
H
−3
, we might also at
tribute to J
−
the same ﬁnite invariant volume.
7
Thus
our construction would seem to ameliorate the horizon
6
This is apparently the case, for example, in the cyclic model [14],
where if the cyclic behavior is to continue indeﬁnitely into the
future and past, it seems necessary to place cosmological bound
ary conﬁtions on an inﬁnite spacelike surface. This speciﬁcation
accords the same properties to points that are arbitrarily distant,
and causally disconnected.
7
One can also consider other ways of taking a limit to J
−
, but
this one seems most in accord with the symmetries of dS. An
other approach to the problem is to rather consider some dis
tance measure between any two points on J
−
. A dSinvariant
quantity, in units where H = 1, between points P
1
and P
2
in
the embedding space (see Eq.(9) below) is given by D(P
1
, P
2
) =
problem (posed in terms of Hubble volumes within the
boundary condition surface) to approximately the same
degree as does inﬂation beginning in the Planck epoch,
and much better than does inﬂation with an early quasi
FRW phase.
V. THE ELLIPTIC VIEW
The model we have proposed consists of two indistin
guishable regions, each comprising an eternally inﬂating
universe with an AOT (where deﬁned) pointing away
from an inﬁnite null surface which connects the two re
gions. The statistical identity of these universes, along
with lack of a global physical time orientation, suggests
some form of an old idea concerning dS, called “the ellip
tic interpretation”
8
that would identify the two universes.
The idea consists of deeming an event to be rep
resented not by a single point of a spacetime mani
fold, but by a pair of antipodal points (deﬁned below).
This corresponds to a topological identiﬁcation that,
applied to our model, identiﬁes regions I and II, and
maps J
−
onto itself (the R × S
2
manifold J
−
becomes
(R × S
2
)/Z
2
). This identiﬁcation has been subject of
some previous [36, 37, 38, 39] and recent [40] investiga
tions.
Pure dS can be represented as a 4D hyperboloid
−v
2
+w
2
+x
2
+y
2
+z
2
= H
−2
(9)
embedded in 4+1 D Minkowski space with metric
ds
2
= −dv
2
+dw
2
+dx
2
+dy
2
+dz
2
. (10)
The elliptic interpretation consists of identifying each
point P with coordinates (v, w, x, y, z) with its antipode
−P at (−v, −w, −x, −y, −z). In the conformal diagram,
this means that points such as P and −P of Fig. 2 are
physically identiﬁed; the antipodal map A looks like a
vertical reﬂection and a horizontal shift through one half
of the diagram’s horizontal extent (Note that although
this map makes an orbifold of the embedding space, there
−v
1
v
2
+w
1
w
2
+x
1
x
2
+y
1
y
2
+z
1
z
2
[11]. In general −∞< D < ∞,
whereas for two points on J
−
, D ≤ 1; they may thus be con
sidered “close”. The relation between D and geodesic distance is,
unfortunately, not always deﬁned. The points are timelike or null
separated, respectively, for D > 1 or D = 1. For D < 1 points
are connectable by a spacelike geodesic of length cos
−1
(D) < π;
but for D ≤ −1 there is no geodesic connecting P
1
and P
2
, al
though one does connect P
1
to the antipode
¯
P
2
of P
2
because
D(P
1
,
¯
P
2
) = −D(P
1
, P
2
). Under the identiﬁcation of antipodal
points (see below), any P
1
and P
2
are nullseparated or spacelike
separated by a geodesic distance < π/2H.
8
The etymology of this term is obscure to the authors. The el
liptic view applied to spatial sections of dS was discussed ﬁrst
by de Sitter, who preferred it to the (now) conventional “spher
ical” view; and cited a letter from Einstein voicing the same
preference [35]. Antipodally identifying in space and time was
discussed in fond detail by Schr¨ odinger [36].
11
J

J

P
X
P
FIG. 4: Conformal diagram of dS including Killing vectors
(arrows), a point P and its antipode −P, along with their
lightcones. The causal diamond of an observer following a
geodesic “X” (comoving in the global coordinatization (7)) is
the interior of the dark diamond, the antipodal copy of which
is the light diamond.
are no ﬁxed points in the dS hyperboloid, leaving the
identiﬁed space a manifold).
At the classical level, the identiﬁcation can be enforced
by demanding that all ﬁelds in dS are symmetric or anti
symmetric under A, and that all sources have an accom
panying antipodal copy. Parikh et al. [40] have argued
that charge conjugation should be added to A, and that
the combination represents CPT. Indeed, in classical ﬁeld
theories at least, Asymmetry automatically entails op
posite charges at antipodal points. Consider, for exam
ple, a complex scalar ﬁeld, satisfying φ(−P) = φ(P).
Then the global time derivative of the ﬁeld at the an
tipode −P is minus that at P, and hence the charge
density (q/2i)
√
−g(φ
∗
∂
0
φ − φ∂
0
φ
∗
) is opposite. How
ever, in 3+1 D at least, we do not recover the PT part
of their argument. A particle at P with 3momentum
vector p = (p
x
, p
y
, p
z
) has an antipodal copy at −P. By
an argument like that of Parikh et al., parallel trans
porting the copy’s trajectory back to P takes the mo
mentum to (−p
x
, p
y
, p
z
). This looks like parity followed
by a rotation of π about the xaxis. The same proce
dure, however, takes a small displacement x = (x, y, z)
to (x, −y, −z). The diﬀerence in sign arises because the
3momentum suﬀers an “automatic” time reversal under
A, whereas the 3displacement does not. Thus the orbital
angular momentum l = x ×p transforms from (l
x
, l
y
, l
z
)
to (−l
x
, l
y
, l
z
), just as the 3momentum does. But under
parity P, x → −x and p → −p, while under T, x → x
and p → −p. Therefore under PT (along with possi
ble rotations), p and l transform oppositely. Only under
T (with rotations) alone do they transform likewise. So
unfortunately we cannot concur with the beguiling idea
that CPT conjugate events occur at antipodally conju
gate points in an elliptic universe. Rather, we must settle
for a sort of CT conjugation between processes at antipo
dal points.
A second interesting (and favorable) feature of the an
tipodal identiﬁcation, which holds in pure elliptic de Sit
ter space (edS), is as follows. An immortal observer in
dS can only eventually “see” (i.e. be connected to via
a nonspacelike geodesic emanating into the past of the
observer) half of the space, the rest being hidden behind
the observer’s dS event horizon. Under the antipodal
identiﬁcation, however, this “hidden” space is exactly the
same as the “visible” space. Likewise, the space behind
the observer’s particle horizon (the space not reachable
by nonspacelike geodesics emanating from an observer
toward its future) is the same as the space within it. In
this precise sense, edS has no horizons. The notion that
in edS each observer has “full information” about the
space has been a prime motivation for the study of the
space [36, 40].
Along with these appealing attributes however, and
like our cosmological model, edS has some unconven
tional temporal features that may or may not be desir
able. Because A includes time reversal, the spacetime is
nontimeorientable: one cannot continuously divide non
spacelike vectors into two classes which can be labeled
“future” and “past”. Now, one may take the view [36]
that since physics is essentially timereversible, this poses
no fundamental problem. Nontimeorientability does,
however, have implications for quantum mechanics (see
Sec. VA and [38, 39]). In addition, while physics may
be time symmetric, our physical world manifestly is not,
and this must be confronted in a cosmological model.
The identiﬁcation of points near “past” inﬁnity with
those near “future” inﬁnity also raises the specter of
closed timelike curves (CTCs) and their accompanying
paradoxes. The identiﬁcation does not allow any self
intersecting timelike curves in perfect dS because the full
lightcones of P and −P never intersect (the spacetime
obeys the strong causality principle of [41]). For the same
reason, no observer can see both an event and its antipo
dal copy. Note, though, that perturbations of dS that
tend to make the conformal diagram “taller” [42] would
allow timelike curves from a point to traverse the dS hy
perboloid to the point’s antipode.
These temporal features may not be what we are ac
customed to, but perhaps we should not be surprised
that any cosmology based on globally dS spacetime has
an “unconventional” AOT. It is well known, for example,
that while a patch of dS admits a static coordinatization,
it admits no global timelike Killing vector; as shown in
Fig. 4, the Killing vector is timelike only in half of the
diagram, and moreover points toward the top of the di
agram only in half of the region in which it is timelike,
reversing completely between antipodal points. (In fact,
the Killing vector ﬁeld maps onto itself under A).
Our model has much the behavior one would want in
a cosmology based on edS. The physical AOT is deﬁned
only inside the bubbles, which may intersect (and hence
compare time orientations) only within region I or region
II. The two regions are separated by locally dS space
time, where physics is fully time symmetric. Under the
antipodal identiﬁcation, the regions are equated, and the
12
physical AOT is consistent everywhere that it can be
compared by two physical observers. Only an imaginary
observer that could travel “back in time” to leave a bub
ble, pass through J
−
, and encounter another bubble,
could see that its own AOT agreed only with that of one
of the two bubbles. (In fact, it would seem that in a non
timeorientable manifold the physical AOT must either
be undeﬁned in some regions, or must suﬀer a reversal
along some surface. Thus a construction something like
ours may well be necessary in any cosmology based on
edS.)
However, an antipodally identiﬁed version of our model
does not quite share all of the desirable properties of
edS. The bubbles, with a larger curvature radius than
the embedding space, allow the connection of antipodal
(and hence identiﬁed) points by a timelike curve. These
selfintersecting timelike curves (SITCs) are not however
CTCs of the usual grandfather paradox sort. To follow
such a SITC an imaginary observer would, for part of
its journey, have to travel backward in (bubble) time.
Moreover, when the two branches of the SITC meet, they
have opposite time orientation as deﬁned by an aﬃne
parameter along the curve. One might avoid these SITCs
if the bubbles have a smaller curvature radius than the
background space (as discussed below in Sec. VI). In
this case, however, horizons would return, because there
would be regions outside of an observer’s horizon that
are not identiﬁed with any region within the horizon.
While the elliptic interpretation is complicated by the
presence of bubbles, the background space of our model is
pure dS, and does beneﬁt from the elliptic interpretation;
all points on J
−
, for example, would be connectable by
geodesics and have a maximal spacelike geodesic separa
tion π/2H, and the volume of the boundary condition
surface would be halved.
The elliptic interpretation was suggested by a symme
try in the statistical description of the bubbles, and it is
interesting to ponder the connection between this statis
tical symmetry and A. Classically, we may have an en
semble of systems each of which is not Asymmetric, even
while the statistical properties of the ensemble are. With
out bubble nucleation, our universe (dS) is Asymmetric.
The boundary conditions which determine the bubble
distribution are Asymmetric, therefore it seems neces
sarily true that the statistics of the bubble distribution
are Asymmetric. Given quantum mechanics, however,
it is not clear how to relate a single member to the sta
tistical properties (given by the wavefunction) without
bringing in measurement theory, and we will leave the
question aside for future consideration.
A. QFT in edS
The nontimeorientability of edS also makes quantum
ﬁeld theory on the space rather more subtle than in usual
dS. We hope to treat this subject in some detail in a
forthcoming paper (see also [38, 39, 40]). Here we sketch
a brief summary.
Consider a massive free scalar ﬁeld φ(x) obeying some
wave equation, for which we would like to construct a
QFT in edS by deﬁning a Hilbert space of states (in
cluding a vacuum state 0), and the twopoint func
tion 0φ(x)φ(y)0. The latter can be decomposed
into a commutator D(x, y) and an anticommutator (or
“Hadamard”) function G
(1)
(x, y). Under the antipodal
identiﬁcation, we would expect both D and G
(1)
to be
symmetric in some sense under the exchange of x and/or
y with its antipode. How, then, might we deﬁne such a
QFT? There are a number of ways, none of which seem
entirely satisfactory.
First, one might just pick a particular “antipodally
symmetric” vacuum state of full dS. Indeed, taking the
α → ∞ limit of the “αvacua” appropriate for dS [43]
does yield an Asymmetric G
(1)
. However, this does not
have the usual short distance behavior of the Minkowski
2point function. In addition, the commutator D is inde
pendent of the state chosen, and has no antipodal sym
metry. (One might also hope to ﬁnd an Asymmetric
nonvacuum state with the correct short distance behav
ior, but this would still have the wrong commutator.)
A second approach would be to try to build an an
tipodally symmetric vacuum in a full dS background, by
choosing a (global) time coordinate and decomposing the
ﬁelds into global positivefrequency modes that are an
tipodally symmetric. The Fock vacuum would then as
usual be the state destroyed by all annihilation opera
tors. The problem with this approach is that any an
tipodally symmetric mode turns out to have vanishing
KleinGordon norm when integrated over a Cauchy sur
face for all of dS [38], and the Fock construction breaks
down.
A third approach employed in the literature [38, 39,
40] is to abandon the hope of a Fock vacuum, and to
simply enforce the antipodal symmetry at the level of
the Green functions, by writing antipodally symmetrized
versions of the ﬁelds, and computing the resulting two
point functions in terms of the twopoint functions of
unidentiﬁed dS. One problem with this approach is that
it becomes somewhat unclear why the anticommutator
should take this value, as the construction seems to lack
an underlying quantummechanical motivation. Another
problem is that while the anticommutator so obtained
is antipodally symmetric, the same procedure yields a
commutator that vanishes identically.
A fourth approach is to deﬁne the vacuum in terms
of a mode decomposition over only part of dS (such as a
“causal diamond”[11]), where the modes can be consis
tently positive or negative frequency (so that a Fock rep
resentation exists), then provide a prescription for deﬁn
ing correlators between any two points in the space in
terms of correlators within this region. This approach
turns out to be promising; for one or both points in the
causal diamond or its antipodal copy, and for a particular
choice of state (a mixed thermal one at twice the usual
de Sitter Hawking temperature), both D and G
(1)
can be
13
made to have the correct symmetry. The problem arises
when both points are outside of the causal diamond and
its copy; in this case the commutator turns out to van
ish for timelike separated points, and not for spacelike
separated points. It is unclear whether this makes sense.
In short, deﬁning a satisfactory QFT in edS is rather
diﬃcult; the diﬃculties stem primarily from the commu
tator function, because it is not symmetric under time
reﬂection, while the anticommutator is. It is then diﬃ
cult for both functions to be symmetric under A, which
includes timereﬂection. It is possible [40] that string
theory in edS will make more sense than in dS. It is also
conceivable that the elliptic view could emerge from a
correct quantum treatment of dS, and that the described
troubles stem from doing QFT on a ﬁxed background not
included in the dynamics. This would be an interesting
issue to pursue in string theory or other theories of quan
tum gravity. For now we must leave it there, and return
to eternal inﬂation.
VI. GENERALIZATIONS AND EXTENSIONS
We have constructed our eternally inﬂating universe
using an “open inﬂation” doublewell potential and de
manding that the inﬂaton φ rest in the false vacuum
everywhere on the inﬁnite null surface J
−
. But these
choices are not unique, and the general principles of our
construction can be extended to models employing diﬀer
ent potentials, or diﬀerent boundary condition surfaces.
A. Diﬀerent inﬂaton behaviors
A simple change in our model can be induced by leav
ing V (φ) ﬁxed, but demanding that φ rest in the true,
rather than false, vacuum on J
−
. Although the ﬁeld is
now in a (positive) stable vacuum, bubbles of false vac
uum may still be able to nucleate [44]. Assuming this
indeed occurs, the eﬀect of each bubble can be enclosed
within a lightcone, and the distribution of these light
cones is essentially the same as for bubbles of true vac
uum, so all of the arguments of Sec II go through. Now,
within these bubbles of false vacuum new bubbles of true
vacuum can form, one of which could describe our ob
servable universe. At late (bubble) times each bubble
interior would approach dS, dominated by the true vac
uum (presumably of the magnitude we currently appear
to observe), in which false vacuum bubbles may nucleate
(and so on, adinﬁnitum). This is a realization of the
“recycling universe” of [45]. Because bubbles are nested
inﬁnitely deep, the structure of this universe appears ex
ceedingly complicated. But the global structure at the
outermost level of bubbles is known (and is as in Fig. 2),
because we have speciﬁed it using explicit cosmological
boundary conditions. One might argue that this scenario
would have yet simpler boundary conditions than the sce
nario in which the inﬂaton is placed in the false vacuum
on J
−
. Since the lowest vacuum state must be posi
tive for this scenario to work, it also has the potential
to connect with the presently observed positive vacuum
energy.
A somewhat similar scenario can be realized using
“chaotic inﬂation” potentials such as V (φ) = λφ
4
+ V
0
.
Here we set the inﬂaton to rest at φ = 0 on J
−
, and
require V
0
> 0 so that the space is locally dS near J
−
.
This is, again, a stable conﬁguration, yet regions of large
potential may “nucleate” in this dS background as highly
improbable quantum ﬂuctuations in the ﬁeld. If such a
region nucleates at high enough φ, it would provide the
potential seed for the sort of eternal inﬂation envisioned
by Ref. [46], in which quantum ﬂuctuations in φ domi
nate over classical rolling so that inﬂation become eter
nal. While the structure of the resulting region becomes
extremely complicated, it can again be contained within
a lightcone of some point in the original (dS) embedding
space, so that the global structure of the universe (ﬁlled
with such lightcones) is still understandable, and again
looks like Fig. 2.
B. Diﬀerent boundaries
We were led to the null surface J
−
by constructing an
eternal model with a timetranslation symmetry. But the
same sort of boundary conditions we apply on J
−
could
also be applied to a spacelike initial surface, such as the
˜
t = 0 surface in the global coordinatization (7) of dS.
Such boundary conditions might be closely tied in to the
HartleHawking NBP, as discussed above. The universe
now has a preferred (and initial) time, and constitutes
“semieternal” inﬂation in both the
˜
t < 0 and
˜
t > 0
regions. The same arguments concerning the AOT apply
here: it is undeﬁned near
˜
t = 0 (outside all bubbles),
and deﬁned in bubbles, pointing away from
˜
t = 0. If the
initial surface maps onto itself under the antipodal map,
we can apply the antipodal identiﬁcation to the universe.
This model bears a stronger resemblance to the earlier
ideas of Sakharov [30, 31, 32, 33] than does our proposal
using the null J
−
.
Spacelike surfaces are easily deformable into other
spacelike surfaces, whereas the same is not true for null
surfaces (see e.g. [47]). Thus spacelike surfaces are in
some sense less constrained than null ones and thus
maybe less appropriate in any attempt, such as ours,
to specify cosmological boundary conditions in the most
economical way. In addition, the resulting universe would
not obey the Perfect Cosmological Principle, having a
preferred time at t = 0. As discussed above, we also
might conjecture that the quantum state corresponding
to null boundary conditions is simpler. In general, how
ever, there seems to be no strong argument against such
a boundary condition surface as compared to a null sur
face.
14
VII. SUMMARY AND CONCLUSIONS
We have investigated the possibility of making “future
eternal” inﬂation eternal also to the past. Starting with
a de Sitter spacetime background dominated by the false
vacuum energy in which truevacuum bubbles can form,
we specify that the bubble distribution at any time is
in exactly the steadystate conﬁguration asymptotically
approached by semieternal inﬂation. All bubbles are
then equivalent, and the statistical distribution of bub
bles admits both a timetranslation and spacetranslation
invariance in the background inﬂating space.
The described region (“region I”) has no initial time,
but has a null boundary J
−
that is the limiting surface
as t → −∞ of the ﬂat spatial sections. The steady
state conﬁguration at t > −∞ corresponds to the inﬂa
ton ﬁeld being in the false vacuum state everywhere on
J
−
, with no incoming radiation from J
−
. This sur
face can be reached by a pastdirected geodesic of ﬁnite
proper length from any point in the space, so region I is
geodesically incomplete; this is the “singularity” pointed
to by theorems purporting that pasteternal inﬂation is
impossible [3, 4]. But the space can (and should) be ex
tended, as the state on J
−
also constitutes boundary con
ditions for the region past J
−
if the manifold is extended.
These boundary conditions imply that a duplicate copy
of region I exists on the other side of J
−
. Together re
gion I and the new “region II” constitute a geodesically
complete (i.e. nonsingular) inﬂating cosmology that ad
mits a coordinatization in which the background space
and bubble distribution are timetranslation and space
translation invariant.
The null surface J
−
constitutes a cosmological bound
ary condition surface on which the universe is classically
in a particularly simple and symmetric state. Although
J
−
is an inﬁnite null surface, some of its points are in
causal contact, and one might attribute a ﬁnite invariant
volume of O(H
−3
) to it. Therefore specifying boundary
conditions on it is not like specifying them on an inﬁ
nite spacelike hypersurface (which would lead to a severe
“horizon problem”).
It might be possible to construct a quantum state cor
responding to our classical boundary conditions on J
−
by
taking a nulllimit of spacelike sections on which the wave
functional describing the ﬁelds is centered on the desired
classical state. But an explicitly null quantum formula
tion of our null boundary proposal has not been provided
and would constitute an interesting future study.
It is widely thought that the “arrow of time” is con
nected with cosmological boundary conditions. In our
model, time ﬂows away from J
−
, and the AOT is con
sistent among all physical observers that can compare it.
The AOT is not, however, deﬁned globally, and in our
model the statistical description of the universe admits a
global symmetry that includes timereversal.
This symmetry, along with the presence of two dupli
cate, noncommunicating universes, motivates—though
does not require—formally identifying antipodal points
on the manifold. We, and others, have studied this iden
tiﬁcation on de Sitter space classically and quantum
mechanically. Antipodally identiﬁed (or “elliptic”) dS
has the virtues that it is causally stable and observers
have no event horizons. It has the disadvantage that its
nontimeorientability makes deﬁning a reasonable quan
tum ﬁeld theory diﬃcult. With antipodal identiﬁcation
our model is more economical as the two duplicate uni
verses are identiﬁed; however not all of the attractive
features of “pure” edS remain.
Our model can be generalized to other inﬂaton po
tentials (such as for chaotic inﬂation), and therefore al
lows one to partially understand the global structure of
the eternally inﬂating spaces that result. One may also
use an analogous construction to specify a semieternally
inﬂating but nonsingular universe by placing boundary
conditions on a spacelike section.
Our primary conclusion is that it is possible, using only
“standard” ingredients underlying popular models for in
ﬂation, to specify simple cosmological boundary condi
tions on an inﬁnite null surface that lead to past and
futureeternal inﬂation. Such a universe would obey the
same Perfect Cosmological Principle that governs a semi
eternally inﬂating universe long after its beginning. If our
construction survives scrutiny, and can be speciﬁed at the
fully quantum (or quantumgravitational) level within a
theory of fundamental physics, it could serve as the basis
for a realistic cosmology that avoids a cosmological sin
gularity, a beginning of time, or a creation of the universe
“from nothing”.
Acknowledgments
We thank Alex Vilenkin for helpful comments on a
draft of this paper. AA is supported by a grant from the
W.M. Keck foundation. SG is supported in part by US
Department of Energy grant DEFG0291ER40671.
APPENDIX
Here we review Green functions and the initial value
problem in some detail, focusing for simplicity on the
inhomogeneous scalar wave equation in ﬁxed background
spacetimes. We are interested in how much information
must be speciﬁed, and where, in order to ﬁx the ﬁeld
throughout spacetime.
1. Deﬁning Green functions
Imagine that our scalar ﬁeld φ satisﬁes
−m
2
φ = q (x
µ
) , (11)
where q is an arbitrary source term, and we wish to de
termine φ at some point P with coordinates x
P
. This is
possible given φ and its time derivative on some complete
spacelike slice through the past lightcone (PLC) of P.
Let us see why this is so, and whether this is the only
suitable set of initial data for the problem.
15
V
S
S
L
FIG. 5: Possible choices of V for diﬀerent Green functions.
The solid lines indicate the regions S of ∂V which contribute
to the integral. On the left, we have volumes suitable for
the retarded and advanced Green functions GR and GA, and
on the right we have volumes suitable for the commutator
function GC.
First introduce another function G, which depends on
P and is assumed to satisfy
−m
2
G = s (x
µ
) , (12)
where s is to be chosen. Taking G times (11) minus φ
times (12) and integrating through some four dimensional
volume V yields
d
4
x∂
µ
G
√
−gg
µν
∂
ν
φ −φ
√
−gg
µν
∂
ν
G
=
√
−g (qG−sφ) d
4
x. (13)
This allows determination of φ(P) in terms of its values
elsewhere by a suitable choice of s, G, and V.
To eﬀect this, we must isolate φ(P), either on the left
or right hand side of (13). Let us ﬁrst use the RHS. Take
s to be a δfunction at P (or, more carefully, a function
peaked near P that can be taken to a δfunction limit).
Then choosing V to enclose P gives φ(P) for the second
term on the RHS of Eq. (13). Using Gauss’s law, this
equation can now be used to express φ(P) as a surface
integral over of φ and its derivative over the boundary ∂V
of V, plus a volume integral over V of the source alone.
To ﬁx G(x), we must choose boundary conditions spe
ciﬁc to our choice of s. Two conventional choices are to
require G to vanish either to the future or to the past of
P. In the ﬁrst case G is known as the “retarded” Green
function and may be denoted G
R
; in the second case it
is “advanced” (G
A
).
We must now choose V (see Fig. 5). It turns out (see,
e.g., [15]) that G
A
= 0 only on and within the FLC of
P, and that G
R
= 0 only on and within the PLC of P.
Hence only a segment S of ∂V contributes to Eq. (13),
enabling us to deduce φ(P) using only data on some con
nected surface making a complete span through either the
FLC or the PLC of P. The usual choice, consistent with
our standard ideas about causality, is to pick a spacelike
surface of constant time (see left side of Fig. 5). Then
evaluation of Eq. (13) requires φ and
˙
φ on S. For a more
general ∂V , we require φ and its normal derivative on S.
(Note that this surface need not be everywhere spacelike,
but if not then the data must be selfconsistent.) The
case of present interest is that of a null surface, for which
the normal lies everywhere within the surface itself (this
is made possible by the Lorentzian signature of the space
time.) Then specifying φ on the surface also speciﬁes the
normal derivative of φ on it. A suitable null surface is,
for example, the forward lightcone L of a point Q to the
past of P; now we need only know φ on the surface S
where this FLC of Q is within the PLC of P (see right
side of Fig. 5). (If the data near Q is suitably regular, the
nonsmooth nature of the surface at Q is unimportant.)
In the same spirit, one might consider a “wedge” such as
the boundary of the future of the segment of a line in the
past of P.
Let us now consider the second way of isolating φ(P),
using the LHS of Eq. (13). Set s = 0, and choose some
piece of ∂V to be spacelike and run through P. Then
choose G to vanish everywhere on this surface. Further,
choose it to have a δfunction at P in its normal deriva
tive; the integral then picks up a contribution propor
tional to φ(P). This choice comprises boundary condi
tions for G, and the function they determine, which we
denote G
C
, now vanishes outside the (full) lightcone of P
and satisﬁes the homogeneous version of Eq. (12). Choos
ing now the remainder of ∂V to close oﬀ to the future or
to the past of P, we again may determine φ(P) once data
is given on a complete slice S through one of the light
cones. G
C
may be called the commutator function, be
cause it turns out to be equal to −i times the commutator
of a free quantum ﬁeld (at least in a globally hyperbolic
spacetime). Note also that G
C
= G
A
− G
R
: subtract
ing the (inhomogeneous) equations governing G
R
and G
A
leaves a homogeneous equation for the diﬀerence. Then
the diﬀerence must vanish outside the lightcone of P,
just as for G
C
.
In all cases, G depends on the point P, and so may
be considered as a function of two variables, x and x
P
.
Because φ satisﬁes (11), it turns out that the G also
satisﬁes (11) with respect to x
P
, at least away from x
P
=
x.
2. Green functions for Minkowski and dS
spacetimes
Let us now outline a procedure to obtain some Green
functions for massive scalar ﬁelds in certain spacetimes,
using a slightly unconventional but perhaps more intu
itive method not relying on the usual Fourier techniques
or iǫprescriptions. We start with the commutator func
tion for a massless ﬁeld in 3+1 D Minkowski space,
G
M0
C
= sgn(t) δ
t
2
−r
2
/2π, (14)
16
with coordinates such that P is at the origin. That
this solves φ = 0 can be seen by expanding out the
δfunction and comparing to a general superposition of
incoming and outgoing spherically symmetric waves. (A
nice discussion of this function is found in [48].) Note
that this is only nonzero on the lightcone itself, vanish
ing even inside the cone. This is a special property of
massless ﬁelds in evendimensional spacetimes (see [15]
for more details.)
To obtain the solution for the massive ﬁeld in
Minkowski space, we start by considering
G
1
≡ f
t
2
−r
2
sgn (t) Θ
t
2
−r
2
.
The sgn and Θ functions enforce the gross features of
the behaviour that we require. Writing using spherical
polar coordinates, we ﬁnd
( −m
2
)G
1
= sgn(t) Θ
t
2
−r
2
−m
2
f
− 4f (0) sgn(t)δ
t
2
−r
2
. (15)
Let us choose f(τ
2
) (where here τ
2
= t
2
− r
2
) such that
−m
2
f = 0 inside the lightcones; a general solution
is f(τ
2
) = AJ
1
(mτ)/τ +BY
1
(mτ)/τ where J
1
and Y
1
are
Bessel functions as in Ref. [49]. The ﬁrst term of Eq. (15)
is then zero. For the second term, we notice that it is
basically the solution to the massless problem that we’ve
already found. So let us simply add the massless solution
G
M0
C
from Eqn. (14) to our ansatz G
1
. Since G
M0
C
= 0,
we have left −m
2
G
M0
C
on the RHS. By choosing A =
−m/4π and B = 0, we obtain a complete cancellation,
leaving us with our desired solution. Thus the full Green
function is
G
M
C
=
sgn (t)
2π
δ
τ
2
−
mΘ(τ
2
)J
1
(mτ)
2τ
. (16)
We may now consider generalizing Eq. (16) to other
spacetimes such as dS. First, let us note that we took
f above to be a function of the proper time τ from the
origin, within the lightcone. Thus the commutator func
tion was invariant under Lorentz boosts. This suggests
that in dS, for example, we make our commutator inside
the forward lightcone invariant under isometries which
leave P ﬁxed. This is the same as saying that it should be
a function of the proper time from the origin alone. One
can deﬁne this quantity in terms of an “angle” in the 5D
embedding space [11] (see footnote to Sec IVD2), but
perhaps a more intuitive way to proceed is as follows. A
patch of dS can be covered by coordinates in which the
metric is the same as that for an open expanding FRW
universe, with scale factor H
−1
sinh Ht. The proper time
along the geodesics representing comoving observers is
just given by the coordinate time t. These geodesics all
intersect at a point as t → 0, which we choose to be
P, and they cover all of the interior of the FLC of P.
We therefore need only ﬁnd an appropriate spatially ho
mogeneous solution of the massive wave equation in the
coordinate patch of the open slicing of dS. The equation
reads:
1
sinh
3
Ht
∂
t
sinh
3
Ht∂
t
φ
+m
2
φ = 0. (17)
To solve this [50], write φ = Hχ/ sinhHt and set z =
coshHt to obtain
(1 −z
2
)χ
,zz
−2zχ
,z
+
2 −
m
2
H
2
−
1
1 −z
2
χ = 0, (18)
which is Legendre’s equation [49] with ν = −1/2 +
9/4 −m
2
/H
2
and µ = −1. The solution to this equa
tion which is regular as z → 1 is P
µ
ν
(z). In terms of t,
this tends to Ht/2 as t →0 independent of m
2
, hence φ
tends to H/2.
We can now use this result to deduce the form of the
commutator over all of dS space. Extend the meaning of
τ
2
to be the signed geodesic distance squared between P
and the point in question. Note that we do not consider
the point to be in the lightcone of the antipode of P,
since no geodesic exists which connects it to P. In this
case we rather deﬁne G
C
to be zero (but see our discus
sion of the antipodal quantum commutator above.) Near
P the space is locally Minkowskian, so we can compare
to our Minkowksi result to obtain:
G
dS
C
=
sgn(t)
2π
¸
δ(τ
2
) −
m
2
Θ(H
2
τ
2
)
2 sinhHτ
P
−1
ν
(coshHτ)
.
(19)
Here by t is meant some suitable generalization of
Minkowski time, with sgn(t) thus serving to make G
dS
C
be of opposite sign in the future and past lightcones, and
τ = +
√
τ
2
. The sgn(t)δ(τ
2
) should be interpreted as the
generalization of the equivalent term in the Minkowski
result.
3. Domains of dependence
Given our Green functions, we would like to investigate
to what extent ﬁxing the ﬁelds on a nullcone L of a
point Q determines the ﬁeld values within that cone. We
are particularly interested in the importance of the ﬁeld
near Q, as our cosmological boundary surface J
−
can
be considered the lightcone with Q “at” pasttimelike
inﬁnity.
Take, as a ﬁrst example, the Green function G
R
for a
massless ﬁeld such as the scalar ﬁeld in 4D Minkowski
space with G
R
from Eq. (14). Because G
R
has support
only on the PLC of P, the ﬁeld at P depends only on the
intersection of P’s PLC and L. This indicates two things.
First, specifying data on L explicitly determines the ﬁeld
everywhere inside L (i.e. everywhere in the future of Q).
Second, the ﬁeld at any point P inside L does not de
pend on φ(Q); we might thus consider Q to be irrelevant
in terms of what occurs within L. This, however, hides
17
a subtlety: while G
r
and φ on L always allow the con
struction of a valid solution of the ﬁeld equations, there
is no guarantee that the ﬁeld so obtained is continuous
with the ﬁeld speciﬁed on L unless some assumptions of
ﬁeld regularity at the vertex (Q) are made: if we wish
to evaluate φ on L to check that we have a genuine solu
tion to our boundary value problem, then we must either
know φ(Q), or assume that the ﬁeld is regular as Q is
approached along L, so that we can extend the ﬁeld to
Q by continuity.
For a massive ﬁeld the strict “Huygen’s Principle” does
not hold; while much of the contribution to the integral
in Eq.(13) comes from the PLC of P where G
R
is singular
(see [15] for some discussion of this “generalized Huygen’s
principle”), there is also a contribution from inside P’s
PLC because G
R
is everywhere nonzero there.
In the dS case, G
R
as given by Eq. (19) falls oﬀ expo
nentially as τ →∞; thus we may expect that when L is
taken to be J
−
, where the vertex lies “at” τ = ∞, the
ﬁeld values inside J
−
will not depend on the ﬁeld at the
vertex (where by “at the vertex” we mean within J
−
in
the limit τ → ∞). More explicitly, we may consider a
boundary surface comprised of J
−
at
˜
t >
˜
t
0
for some
˜
t
0
, closed by the spacelike surface
˜
t =
˜
t
0
, where
˜
t and
˜
t
0
are in the global closed coordinates. Then the ﬁeld
integral (13) consists of an integral along J
−
, and an
integral over the
˜
t =
˜
t
0
hypersurface, which has a ﬁnite
physical volume of order H
−3
. But then as
˜
t
0
→ −∞,
τ → ∞, so this second contribution will vanish unless
either the average of the ﬁeld or its
˜
tderivative blow up
faster than G
−1
R
. Hence we expect that the ﬁeld at any
point in the space will depend only on the values on J
−
,
and not on the ﬁeld behavior at past inﬁnity, as long as
the ﬁelds are assumed to be suitably ﬁnite and regular.
This argument might also be applicable to gravity; the
equations governing the Weyl conformal tensor can be
cast, at the linear level, as waveequations for a spin2
ﬁeld, and boundary conditions can be speciﬁed on a null
surface such as J
−
[24]. We therefore expect that our
boundary conditions on J
−
determine the Weyl tensor
uniquely, and therefore determine the spacetime to be
pure dS when bubble nucleations are neglected.
[1] A. Linde, D. Linde and A. Mezhlumian, Phys. Rev. D
54, 2504 (1996); A. Vilenkin, Phys. Rev. D 46, 2355
(1992).
[2] See, for example, A.H. Guth, Phys. Rept. 333, 555 (2000)
and references therein; A.D. Linde, Phys. Lett. B175,
395 (1986); A.D. Linde in The Very Early Universe,
edited by G.W. Gibbons, S.W. Hawking and S.T.C. Sik
los (Cambridge University Press, Cambridge, England,
1983); P.J. Steinhardt, same volume.
[3] For a review, see A. Borde and A. Vilenkin, Int. J. Mod.
Phys. D5 (1996) 813824.
[4] A. Borde, A.H. Guth and A. Vilenkin, grqc/0110012.
[5] A. Aguirre and S. Gratton, Phys. Rev. D 65 (2002)
083507.
[6] A.H. Guth, Phys. Rev. D 23, 347 (1981).
[7] M. Bucher, A.S. Goldhaber and N. Turok, Phys. Rev. D
52, 3314 (1995).
[8] J.R. Gott, Nature (London) 295, 304 (1982); K. Ya
mamoto, M. Sasaki, and T. Tanaka, Astrophys. J. 455,
412 (1995); A. Linde and A. Mezhlumian, Phys. Rev. D
52, 6789 (1995)
[9] A. Vilenkin, Phys. Rev. D 46, 2355 (1992).
[10] A.H. Guth and E.J. Weinberg, Nucl. Phys. B212, 321
(1983).
[11] For reviews of de Sitter space, see: Ref. [41]; M. Spradlin,
A. Strominger, and A. Volovich, hepth/0110007.
[12] S. Coleman and F. De Luccia, Phys. Rev. D 21, 3305
(1980).
[13] B. Mandelbrot, The Fractal Geometry of Nature, (W.H.
Freeman and Co., New York, 1983).
[14] P.J. Steinhardt and N. Turok, hepth/0111030.
[15] R. Courant, Methods of Mathematical Physics: Vol. II,
Partial Diﬀerential Equations (New York: Interscience
Publishers, John Wiley & Sons, 1962).
[16] R. Penrose, Gen. Rel. and Grav. 12, 225 (1980).
[17] V.A.Rubakov and S.M.Sibiryakov Theor. Math. Phys.
120, 1194 (1999).
[18] See, e.g., H.D. Zeh, “The physical basis of the direction
of time”, (Berlin: SpringerVerlag, 1999).
[19] R. Penrose, in “Quantum Gravity 2: A Second Oxford
Symposium”, ed. C.J. Isham, et al. (Oxford: Clarenden
Press), 1981.
[20] G.F.R. Ellis and R. Maartens, qrqc/0211082.
[21] R. Penrose, in “Fourteenth Texas Symposium on Rel
ativistic Astrophysics, Dallas, Tex., 1988” (New York:
Annals of the New York Academy of Sciences, 1989); S.
Hollands and R. Wald; grqc/0205058; N. Turok, Class.
Quantum Grav. 19, 3449 (2002).
[22] A. Linde, “Particle physics and inﬂationary cosmology”,
(Switzerland: Harwood Acedemic Publishers, 1990).
[23] T. Vachaspati and M. Trodden, Phys. Rev. D 61, 023502
(2000).
[24] R. Penrose and W. Rindler, “Spinors and SpaceTime,
Vol. I”, (Cambridge: Cambridge University Press, 1984).
[25] T. Heinzl, Lect. Notes Phys. 572, 55 (2001),
hepth/0008096.
[26] S. Gratton and N. Turok, Phys. Rev. D 63, 123514
(2001).
[27] S.W. Hawking and T. Hertog, Phys. Rev. D 66, 123509
(2002).
[28] S.W. Hawking and R. Penrose, The Nature of Space and
Time (Princeton University Press, Princeton, New Jer
sey, USA, 1996).
[29] A. Vilenkin, Phys. Rev. D 27, 2848 (1983).
[30] A. D. Sakharov, JETP Lett. 5, 24 (1967).
[31] A. D. Sakharov, Sov. Phys. JETP 52, 349 (1980).
[32] A. D. Sakharov, Sov. Phys. JETP 56, 705 (1982).
[33] A. D. Sakharov, Sov. Phys. JETP 60, 214 (1984).
[34] A. H. Guth, Phys. Rev. D 23, 347 (1981).
[35] W. de Sitter, M.N.R.A.S 78,3 (1917).
[36] E. Schr¨odinger, Expanding Universes (Cambridge Uni
versity Press, Cambridge, England, 1956).
[37] H. Schmidt, Fortschr. Phys. 41, 179 (1993).
[38] G. Gibbons, Nuc. Phys. B271, 497 (1986).
18
[39] N. Sanchez, Nuc. Phys. B294, 1111 (1987); J.L. Fried
man and A. Higuchi, Phys. Rev. D 52, 5687 (1995).
[40] M. Parikh, I. Savonije and E. Verlinde, hepth/0209120.
[41] S.W. Hawking and G.F.R. Ellis, The large scale structure
of spacetime (Cambridge University Press, Cambridge,
England, 1973).
[42] F. Leblond, D. Marolf and R. C. Myers, JHEP 0206, 052
(2002).
[43] E. Mottola, Phys. Rev. D 31, 754 (1985); B. Allen, Phys.
Rev. D 32, 3136 (1985).
[44] K. Lee and E.J. Weinberg, Phys. Rev. D 36, 1088 (1987).
[45] J. Garriga and A. Vilenkin 1998, Phys. Rev. D 57, 2230
(1998).
[46] A. Linde, D. Linde and A. Mezhlumian, Phys. Rev. D
49, 1783 (1994).
[47] R. Bousso, Rev. Mod. Phys. 74, 825 (2002).
[48] P.A.M. Dirac, Lectures on Quantum Field Theory, (New
York: Belfer Graduate School of Science), 1996.
[49] M. Abramowitz and I. Stegun, “Handbook of Mathemat
ical Functions” (New York: Dover Publications, 1972).
[50] M. Sasaki, T. Tanaka and K. Yamamoto, Phys. Rev. D
51, 2979 (1995).
This action might not be possible to undo. Are you sure you want to continue?